Intelligent wearing state monitoring method and system of ai-based protective equipment

Abstract

This application provides an AI-based intelligent wearable status monitoring method and system for protective equipment, relating to the field of data processing technology. The method includes: monitoring and acquiring a set of workshop environmental parameter sequences within a preset time range in a chemical production workshop; acquiring mask attribute information of a gas mask worn by a user to be monitored; collecting a sequence of user facial images to obtain a mask-face pressure distribution sequence of the gas mask; performing mask-face fit analysis on the user based on the user facial image sequence and the mask-face pressure distribution sequence, and outputting the mask-face fit degree; predicting the mask failure risk of the gas mask based on the set of workshop environmental parameter sequences, mask attribute information, and mask-face fit degree, and outputting the mask failure probability, thereby assessing and issuing early warnings regarding the protective risks to the user. This solves the technical problem of insufficient accuracy and reliability in gas mask wearing status monitoring in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to a method and system for monitoring the status of intelligent wearable protective equipment based on AI. Background Technology

[0002] Chemical plant production environments contain high-risk factors such as toxic gases and harmful particulate matter, and environmental parameters often fluctuate dynamically, placing extremely high demands on the protective performance and wearing status of gas masks.

[0003] However, traditional methods for monitoring the wearing status of gas masks usually focus only on the single dimension of whether they are worn, without incorporating dynamic environmental changes and the properties of the masks themselves into the monitoring and evaluation criteria. This makes it impossible to determine whether the protective performance of the equipment matches the current environmental requirements, resulting in insufficient monitoring accuracy and reliability. This increases the probability of workers being exposed to hazardous environments and becomes a significant hidden danger to chemical safety production.

[0004] Therefore, there is an urgent need for an intelligent wearable status monitoring method that combines AI to accurately determine the suitability of gas mask protection performance with the current environment and provide early warning of failure risks. Summary of the Invention

[0005] This invention addresses the technical problem of insufficient accuracy and reliability in monitoring the wearing status of gas masks in existing technologies by providing an AI-based intelligent wear status monitoring method and system for protective equipment.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0007] In a first aspect, the present invention provides an AI-based intelligent wearable status monitoring method for protective equipment, comprising:

[0008] The system monitors and acquires a set of workshop environmental parameter sequences within a preset time range for chemical production workshops.

[0009] Obtain the mask attribute information of the gas mask worn by the user to be monitored;

[0010] The user's facial image sequence is acquired by an industrial camera, and the mask-face pressure distribution sequence of the gas mask is acquired by a pressure sensor.

[0011] Based on the user's facial image sequence and mask-face pressure distribution sequence, a mask-face fit analysis is performed on the user to be monitored, and the mask-face fit degree is output.

[0012] Based on the set of workshop environmental parameters, mask attribute information, and mask-face fit, the risk of mask failure of the gas mask is predicted, the mask failure probability is output, and the protection risk of the monitored user is judged and warned based on the mask failure probability.

[0013] Secondly, the present invention provides an AI-based intelligent wearable status monitoring system for protective equipment, comprising:

[0014] The environmental parameter acquisition module is used to monitor and acquire a set of workshop environmental parameter sequences within a preset time range in the chemical production workshop;

[0015] The mask attribute acquisition module is used to acquire the mask attribute information of the gas mask worn by the user to be monitored;

[0016] Wearing a data acquisition module is used to acquire a sequence of facial images of the user to be monitored through an industrial camera, and to acquire a sequence of mask-face pressure distribution of the gas mask through a pressure sensor;

[0017] The fit analysis module is used to perform mask-face fit analysis on the user to be monitored based on the user's facial image sequence and mask-face pressure distribution sequence, and output the mask-face fit.

[0018] The risk warning module is used to predict the risk of mask failure of the gas mask based on the set of workshop environmental parameter sequences, mask attribute information and mask-face fit, output the mask failure probability, and make a protection risk judgment and warning for the user to be monitored based on the mask failure probability.

[0019] The beneficial effects of this invention are:

[0020] Compared to existing technologies, this application first monitors and acquires a set of workshop environmental parameter sequences within a preset time range in a chemical production workshop, providing a reliable environmental data foundation for subsequent failure risk prediction. Secondly, it acquires the mask attribute information of the gas masks worn by the monitored users, obtaining a self-portrait of the gas masks and providing key data on the mask's performance for subsequent failure probability prediction. Thirdly, it collects user facial image sequences using industrial cameras and acquires mask-face pressure distribution sequences using pressure sensors, providing a reliable data foundation for subsequent mask-face fit analysis. Furthermore, based on the user facial image sequences and mask-face pressure distribution sequences, it performs mask-face fit analysis on the monitored users, outputting the mask-face fit degree, achieving a quantitative assessment of mask-face fit, adapting to the complex operating scenarios of chemical workshops, and providing reliable fit status data for subsequent failure risk prediction. Finally, based on the set of workshop environmental parameter sequences, mask attribute information, and mask-face fit, the mask failure risk of the gas mask is predicted, and the mask failure probability is output. Based on the mask failure probability, the protection risk assessment and early warning for the monitored users are performed. The mask failure risk prediction plugin with multiple model combinations is used to deal with the complexity of chemical scenarios, and the prediction accuracy and efficiency are balanced by dynamically selecting the number of models. In the end, reliable, efficient and accurate failure risk prediction is achieved, providing a scientific basis for protection risk assessment and early warning, and reducing the probability of employees being exposed to dangerous environments.

[0021] Through the aforementioned technical solution, this application integrates workshop environmental parameter sequences, mask attribute information, user facial image sequences from a visual dimension, and mask-face pressure distribution sequences from a pressure dimension to achieve comprehensive perception of the protective status. Furthermore, based on a mask failure risk prediction plugin constructed using long short-term memory networks and convolutional neural networks, it predicts and outputs the mask failure probability, enabling real-time capture of risks such as environmental fluctuations, decreased mask-face fit, and performance degradation of gas masks. This improves the accuracy and reliability of protective monitoring, effectively reducing the exposure risk to workers and the probability of chemical safety accidents, providing dynamic and precise safety assurance for high-risk work scenarios. Attached Figure Description

[0022] Figure 1 A flowchart illustrating the AI-based intelligent wearable status monitoring method for protective equipment provided by this invention.

[0023] Figure 2 This is a schematic diagram of the structure of the AI-based intelligent wearable status monitoring system for protective equipment provided by the present invention.

[0024] In the attached diagram, the components represented by each number are as follows:

[0025] Module 11 for acquiring environmental parameters, Module 12 for acquiring mask attributes, Module 13 for collecting wearing data, Module 14 for analyzing fit, and Module 15 for risk warning. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0028] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0029] Example 1, as Figure 1 As shown, this embodiment of the invention provides an AI-based intelligent wearable status monitoring method for protective equipment, including:

[0030] S10: Monitor and acquire a set of workshop environmental parameter sequences within a preset time range for chemical production workshops.

[0031] The environmental parameters of chemical production workshops are highly dynamic, and changes in these parameters are a core external factor that can trigger the failure of gas masks.

[0032] To address the aforementioned issues, this application monitors and obtains a set of workshop environmental parameter sequences within a preset time range for chemical production workshops.

[0033] Specifically, step S10 in the method includes:

[0034] Configure environmental monitoring indicators, including the composition and concentration of toxic gases, the composition and concentration of particulate matter, temperature, humidity and air velocity;

[0035] Based on the environmental monitoring indicators, the environmental parameters of the chemical production workshop are monitored and obtained at multiple continuous monitoring time points within a preset time range, resulting in multiple workshop environmental parameter sequences, and a set of workshop environmental parameter sequences is constructed.

[0036] In this embodiment, environmental monitoring indicators are first configured, including the composition and concentration of toxic gases, the composition and concentration of particulate matter, temperature, humidity, and air velocity. Among these, toxic gases commonly found in chemical production (such as hydrogen sulfide, chlorine, benzene compounds, etc.) are included. The composition of the toxic gases determines the filtration suitability, and the concentration determines the protection load.

[0037] Among them, particulate matter such as dust (such as resin dust), droplets (such as acid mist), and aerosols from chemical production affect filtration efficiency, and excessively high concentrations of particulate matter will accelerate filter clogging.

[0038] High temperatures can cause the rubber sealing ring of the mask to soften, while low temperatures can cause it to harden. Both of these conditions can disrupt the seal between the mask and the face, leading to leaks. Furthermore, the adsorption capacity of some filter media decreases at high temperatures, while the reaction rate slows down at low temperatures, directly affecting the protective effect.

[0039] High humidity can cause filter materials such as activated carbon to become damp, leading to clogged pores and an inability to adsorb toxic gases. In addition, high humidity can cause facial sweating, reducing the friction between the mask and the face and making it prone to shifting. At the same time, sweat may corrode the mask's sealing ring, reducing its sealing performance.

[0040] Excessive airflow velocity can cause toxic gases / particulate matter to come into contact with the mask surface more quickly. If there is a tiny leak in the mask, harmful substances can penetrate the leak point and enter the face more quickly. It may even blow away the edge of the mask, causing the mask to shift, disrupting the original fit, and increasing the risk of leakage.

[0041] Secondly, based on environmental monitoring indicators, workshop environmental parameters are monitored and acquired at multiple continuous monitoring time points within a preset time range in the chemical production workshop, resulting in multiple workshop environmental parameter sequences, which are then used to construct a set of workshop environmental parameter sequences. The preset time range can be set according to the actual operating scenarios of the chemical plant, ensuring that the collected workshop environmental parameters cover the complete time period during which users actually wear masks, avoiding data fragmentation. For example, the time of one work shift can be set as the preset time range. Continuous monitoring time points are data collected within the preset time range at a certain collection frequency (e.g., once every 5 minutes, once every 10 minutes, etc.). For example, for a specific environmental monitoring indicator, the workshop environmental parameters at each continuous monitoring time point are mapped one-to-one with the continuous monitoring time point, forming multiple workshop environmental parameter sequences. Each workshop environmental parameter sequence reflects the changing trend of a single environmental monitoring indicator over time. The workshop environmental parameter sequences of all environmental monitoring indicators are then aggregated to form a set of workshop environmental parameter sequences covering multi-dimensional environmental variables and possessing temporal characteristics.

[0042] In summary, compared to existing technologies, this application monitors and acquires a set of workshop environmental parameter sequences within a preset time range for chemical production workshops. This provides a reliable environmental data foundation for subsequent failure risk prediction.

[0043] S20: Obtain the mask attribute information of the gas mask worn by the user to be monitored.

[0044] The inherent properties of a gas mask are the core intrinsic factors that determine its baseline protective capability and affect the risk of failure. When the inherent properties of a gas mask cannot meet the current protection requirements, it will directly or indirectly increase the probability of gas mask failure and ultimately affect the effectiveness of protection.

[0045] To address the aforementioned issues, this application obtains the mask attribute information of the gas mask worn by the user to be monitored.

[0046] Specifically, step S20 in the method includes:

[0047] The mask attribute information of a gas mask should include at least its structural design features, filtration performance features, applicable environmental features, and remaining service life.

[0048] Among them, structural design features directly affect the fit between the mask and the face, including at least: mask type: such as half mask (only covering the mouth and nose), full mask (covering the entire face); sealing edge design: such as the material and edge shape of the sealing strip; fixing and adjustment structure: such as the number of headbands and the tightness adjustment method, etc.

[0049] Among these factors, filtration performance characteristics determine whether a mask can effectively block harmful substances in the current environment. If the concentration of toxic gases in the workshop exceeds the adsorption capacity of the filter element, even with a 100% fit, harmful substances will still penetrate the filter element and enter the mask. Therefore, filtration performance characteristics are one of the key bases for failure risk monitoring, and at least include: filter element type: such as filter cartridges, filter cotton, etc. Different filter elements are only effective against specific pollutants; filtration efficiency parameters: such as filtration efficiency for particulate matter and adsorption capacity for toxic gases. When the concentration of pollutants in the workshop exceeds the filtration efficiency or adsorption capacity, the mask will lose its filtration capacity; dust holding capacity / saturation threshold: such as the maximum amount of dust that filter cotton can hold and the saturation point of the filter cartridge for adsorbing pollutants.

[0050] Among them, the applicable environmental characteristics are the environmental boundaries when the mask is designed. They can be used to determine whether the current environmental conditions in the workshop are within the safe working range of the mask. If the range is exceeded, the mask's performance will degrade. These include at least: Temperature applicable range: If the workshop temperature exceeds the mask's temperature applicable range, it will cause the rubber sealing edge of the mask to age and harden, lose elasticity, reduce sealing performance, become brittle, and crack; Humidity applicable range: Such as the relative humidity range. If the workshop humidity exceeds the mask's humidity applicable range, it will reduce the adsorption capacity of the filter element and accelerate the rusting of metal parts (such as headband buckles), affecting the stability of the fixed structure.

[0051] Among them, the remaining service life reflects the remaining time from the current state of the mask to its failure, which is the key to determining whether the mask needs to be replaced in time. It includes at least the remaining service life of the filter element, the remaining service life of the mask body, and the remaining service life of the vulnerable parts.

[0052] Thus, structural design features determine the physical seal between the gas mask and the face; filtration performance characteristics determine the gas mask's ability to block specific pollutants; applicable environmental characteristics determine the performance stability boundary of the gas mask in the working environment; and remaining service life determines the upper limit of effective protection duration. When one or more of these attributes fail to meet current protection requirements, it will directly or indirectly increase the probability of gas mask failure, ultimately affecting the effectiveness of protection. Therefore, the inherent properties of the gas mask are a key prerequisite for assessing its potential failure.

[0053] In summary, compared to existing technologies, this application obtains the mask attribute information of the gas mask worn by the user to be monitored. This provides a capability profile of the gas mask, offering crucial data on the mask's performance for subsequent prediction of mask failure probability.

[0054] S30: Acquire the user's facial image sequence using an industrial camera, and obtain the mask-face pressure distribution sequence of the gas mask using a pressure sensor.

[0055] Since workshop employees frequently bend over, turn their heads, and operate equipment during actual work, it is necessary to obtain dynamic change information on the wearing status of gas masks to provide a dynamic data basis for subsequent prediction of mask-face fit.

[0056] To address the aforementioned issues, this application uses an industrial camera to capture a sequence of user facial images of the user to be monitored, and uses a pressure sensor to monitor and obtain a sequence of mask-face pressure distribution of the gas mask.

[0057] Specifically, step S30 in the method includes:

[0058] The user's facial images are acquired by an industrial camera at multiple consecutive monitoring time points within a preset time range, resulting in a sequence of user facial images.

[0059] The mask-face pressure distribution of the gas mask is obtained by monitoring multiple consecutive monitoring time points within a preset time range using a pressure sensor, thus obtaining a mask-face pressure distribution sequence.

[0060] In this embodiment, an industrial camera is first used to capture facial images of the user under monitoring at multiple consecutive monitoring time points within a preset time range, resulting in a sequence of facial images. The preset time range can be set according to the actual work scenario in the workshop, ensuring that the captured facial images cover the entire period during which the user actually wears the mask. For example, the preset time range can be set to the duration of a single employee's work session. For instance, an industrial-grade high-definition camera is deployed in the work area of ​​the target workshop. Within the preset time range (e.g., one hour of a single employee's work session), at a certain acquisition frequency (e.g., one frame every 2 seconds), 1800 facial images of the user at consecutive monitoring time points are captured. These images are then arranged chronologically to form a sequence of facial images. This sequence records the dynamic state of mask wearing and can reflect the contact boundary between the mask and the user's face, such as whether the mask's sealed edge covers the mouth, nose, and cheeks, and whether there are obvious gaps.

[0061] Secondly, the mask-face pressure distribution of the gas mask is acquired by monitoring multiple consecutive monitoring time points within a preset time range using pressure sensors, resulting in a mask-face pressure distribution sequence. For example, several miniature pressure sensors are embedded in the sealing edges of the gas mask (such as around the mouth and nose, and on both sides of the cheeks). Within a preset time range, the pressure values ​​of each miniature pressure sensor are acquired at a frequency synchronized with image acquisition, resulting in mask-face pressure distributions at multiple consecutive monitoring time points. These distributions are then arranged chronologically to form a mask-face pressure distribution sequence. This sequence reflects the actual tightness of contact between the mask and the face, supplementing the user's facial image sequence.

[0062] In summary, compared to existing technologies, this application acquires user facial image sequences of the user to be monitored using an industrial camera, and obtains the mask-face pressure distribution sequence of the gas mask using a pressure sensor. This provides a reliable data foundation for subsequent mask-face fit analysis.

[0063] S40: Perform mask-face fit analysis on the user to be monitored based on the user's facial image sequence and mask-face pressure distribution sequence, and output the mask-face fit degree.

[0064] The visual dimension of user facial image sequence can intuitively capture the spatial coverage state of the mask on the face, while the pressure dimension of mask-face pressure distribution sequence can quantify the physical contact tightness between the mask and the face through pressure. However, user facial image sequence may have false fit misjudgment, and visual judgment alone cannot determine whether the mask and face are actually pressed together. Due to the limited deployment range of pressure sensors, the mask-face pressure distribution sequence cannot monitor the fit state in areas not covered by pressure sensors. Therefore, combining user facial image sequence and mask-face pressure distribution sequence forms a complementary and mutually offsetting synergistic relationship to obtain the mask-face fit degree.

[0065] To address the aforementioned issues, this application performs mask-face fit analysis on the user to be monitored based on the user's facial image sequence and mask-face pressure distribution sequence, and outputs the mask-face fit degree.

[0066] Specifically, step S40 in the method includes:

[0067] Based on the mask-face fit predictor, a first mask-face fit ratio sequence is predicted according to the user's facial image sequence;

[0068] The second mask-face fit ratio sequence was obtained by evaluating the mask-face pressure distribution sequence.

[0069] The mask-face fit is calculated based on the first mask-face fit ratio sequence and the second mask-face fit ratio sequence.

[0070] In this embodiment, a first mask-face fit prediction sequence is first obtained based on a mask-face fit predictor and the user's facial image sequence. The mask-face fit predictor is constructed based on a convolutional neural network and can predict the percentage of fit between the mask and the face based on the user's facial image sequence. For example, if the mask covers 90% of the key protective areas such as the mouth, nose, and cheeks in a certain frame of the user's facial image sequence, then the first mask-face fit ratio is output as 90%. The prediction results of all frames of facial images are sorted by time to form the first mask-face fit ratio sequence, such as [90%, 88%, 89%, ..., 87%]. In this way, the user's facial image sequence is converted into quantitative data of fit.

[0071] Secondly, a second mask-face fit ratio sequence is obtained by evaluating the mask-face pressure distribution sequence. For example, an effective pressure threshold, such as ≥3 kPa, can be pre-set based on historical prior data or experimental data. When the mask-face pressure is lower than the effective pressure threshold, it indicates a gap between the mask and the face. Then, for each time point in the mask-face pressure distribution sequence, the percentage of pressure sensors with mask-face pressure distribution ≥ the effective pressure threshold is calculated as the second mask-face fit ratio. Finally, the second mask-face fit ratios at all time points are sorted by time to obtain the second mask-face fit ratio sequence.

[0072] For example, if a gas mask has 5 built-in pressure sensors, and at a certain time point within the mask-face pressure distribution sequence, the mask-face pressure distribution data of 3 of the pressure sensors are greater than or equal to the effective pressure threshold, then the second mask-face fit ratio at that time point is (3 / 5) × 100% = 60%. The mask-face fit ratio at other time points is calculated using the same method, and finally sorted by time to obtain the second mask-face fit ratio sequence, such as [60%, 62%, 80%, ..., 58%]. In this way, the fit status is quantified based on the mask-face pressure distribution sequence, making up for the lack of visual dimension.

[0073] Finally, since the first mask-face fit ratio sequence based on the visual dimension and the second mask-face fit ratio sequence based on the pressure dimension may be unreliable, the mask-face fit is dynamically calculated based on the first mask-face fit ratio sequence and the second mask-face fit ratio sequence. The mask-face fit is a comprehensive fit index. The higher the mask-face fit, the better the overall fit and the lower the risk of failure.

[0074] Specifically, the phrase "based on a mask-face fit predictor, predicting a first mask-face fit ratio sequence according to the user's facial image sequence" includes:

[0075] Based on historical wearing monitoring records of similar gas masks in chemical production workshops, a set of facial images of sample users was collected, and the mask-face fit ratio of different sample user facial images was labeled to obtain a set of sample fit ratio labels.

[0076] Using the set of sample user facial images as input and the set of sample fitting ratio labels as supervision, a convolutional neural network is trained until convergence to generate a mask-face fitting predictor.

[0077] Using the mask-face fit predictor, a first mask-face fit ratio sequence is predicted based on the user's facial image sequence.

[0078] In this embodiment, firstly, based on historical wearing monitoring records of similar gas masks in chemical production workshops, a large number of sample user facial images covering different face shapes (such as round face, square face, long face), different wearing states (such as fitted, half-fitted, not fitted), and different work scenarios (such as bending over, turning head, stationary) are collected as a sample user facial image set. Then, each sample user facial image is manually labeled with the mask-face fit ratio according to a unified labeling standard to obtain a sample fit ratio label set.

[0079] Secondly, using a set of sample user facial images as input and a set of sample fitting ratio labels as supervision, a convolutional neural network is trained until convergence to generate a mask-face fitting predictor. For example, the mask-face fitting predictor can be obtained through the following technical path: 1. Data preparation: Divide the set of sample user facial images and the corresponding set of sample fitting ratio labels into a training set, a validation set, and a test set in a ratio of 7:1.5:1.5. 2. Model Construction: Since Convolutional Neural Networks (CNNs) excel at extracting local features from images and are better suited to the complex wearing conditions of different face shapes and clothing styles in chemical plant environments than traditional image recognition algorithms, a mask-face fitting predictor can be built based on CNNs. This predictor mainly consists of an input layer, a feature extraction module, a feature fusion and dimensionality reduction module, and an output layer. The input layer receives sample user facial images. The feature extraction module contains three convolutional blocks, each consisting of a convolutional layer with 32 / 64 / 128 3×3 kernels, a ReLU activation function, a 2×2 max pooling layer, and a batch normalization layer. It progressively extracts features from low-level (edges / textures) to high-level (fitting region contours), while normalization mitigates feature shifts caused by lighting fluctuations in the workshop. The feature fusion and dimensionality reduction module flattens the output of the last convolutional block into a one-dimensional vector and connects it to a fully connected layer with 256 neurons. The output layer outputs the predicted fitting ratio through a fully connected layer (linear activation function) with one neuron. 3. Model Training: Using sample user facial images from the training set as input and corresponding sample fitting ratio labels as supervision targets, the Adam optimizer (initial learning rate 0.001, decaying by 10% every 5 rounds) is used to minimize the mean squared error loss function. After each training round, the prediction error (e.g., mean absolute error MAE) is calculated using the validation set. When the change in MAE on the validation set is less than 0.5% for 5 consecutive rounds and the MAE on the test set is ≤3%, the model is considered to have converged, training is stopped, and the mask-face fitting predictor is obtained.

[0080] Finally, the pre-trained mask-face fitting predictor is retrieved, and given the input sequence of user facial images, a first mask-face fitting ratio sequence is predicted. For example,

[0081] Specifically, the phrase "calculating the mask-face fit degree based on the first mask-face fit ratio sequence and the second mask-face fit ratio sequence" includes:

[0082] The mean values ​​of the first mask-face fitting ratio sequence and the second mask-face fitting ratio sequence are calculated respectively to obtain the mean value of the first mask-face fitting ratio and the mean value of the second mask-face fitting ratio.

[0083] The mask contour matching degree is calculated based on the facial physiological characteristics of the user to be monitored and the structural design features of the gas mask;

[0084] The ratio of the mask contour matching degree to the preset standard mask contour matching degree is set as the first weight adjustment coefficient, and multiplied by the first initial weight to obtain the first optimized weight, wherein the first initial weight is 0.5, and the first optimized weight is greater than or equal to 0.3 and less than or equal to 0.7.

[0085] The second optimization weight is obtained by subtracting the first optimization weight from 1.

[0086] Based on the first optimization weight and the second optimization weight, the average value of the first mask-face fit ratio and the average value of the second mask-face fit ratio are weighted and calculated to obtain the mask-face fit degree.

[0087] In this embodiment, the arithmetic mean of the first mask-face fit ratio sequence and the second mask-face fit ratio sequence is calculated to obtain the average value of the first mask-face fit ratio (e.g., 80%) and the average value of the second mask-face fit ratio (e.g., 75%). Calculating the average value can effectively reduce the interference of random errors. The higher the average value of the first mask-face fit ratio, the better the overall visual coverage of the mask and the face. The higher the average value of the second mask-face fit ratio, the tighter the overall physical contact between the mask and the face.

[0088] Secondly, the mask contour matching degree is calculated based on the facial physiological characteristics of the user to be monitored and the structural design features of the gas mask. The mask contour matching degree refers to the percentage match between the user's facial physiological characteristics and the structural design features of the gas mask, reflecting whether the gas mask itself fits the user's face shape. A higher mask contour matching degree indicates a better fit to the user's face shape, meaning a better basic fit. For example, frontal / side images of the user to be monitored can be acquired using an industrial camera, and facial physiological characteristics such as face shape, face width, nose bridge height, and jaw curvature can be extracted from them. Structural design features can also be extracted from the mask attribute information of the gas mask worn by the user. Then, the matching degree of each feature is calculated, and finally, the average of multiple feature matching degrees is calculated to obtain the mask contour matching degree. For example, if the width of the user's face is 12cm and the internal design width of the gas mask is 11.5-12.5cm, the matching degree is 100%; if the height of the user's nose bridge is 3cm and the design support height of the gas mask is 2.8-3.2cm, the matching degree is also 100%. Other feature matching degrees are calculated using the same method, and finally, the average of multiple feature matching degrees is calculated to obtain the mask contour matching degree, such as 97%. A higher mask contour matching degree indicates a better fit, which in turn indicates that the average mask-face fit ratio in the visual dimension is more reliable. Conversely, a lower mask contour matching degree indicates that there may be false coverage in the visual dimension, which in turn indicates that the average mask-face fit ratio in the visual dimension is less reliable.

[0089] Next, the ratio of the mask contour matching degree to the preset standard mask contour matching degree is set as the first weight adjustment coefficient, and multiplied by the first initial weight to obtain the first optimized weight. The preset standard mask contour matching degree can be dynamically set based on historical prior data, industry-recognized high-fit standards, and actual usage scenarios; for example, 90%. The first optimized weight = (mask contour matching degree / preset standard mask contour matching degree) × first initial weight. The first initial weight is 0.5, meaning that the average first mask-face fit ratio and the average second mask-face fit ratio are considered equally important. The first optimized weight is greater than or equal to 0.3 and less than or equal to 0.7 to avoid excessively high weights for either the first or second mask-face fit ratio, which could lead to extreme results. When the calculated first optimized weight exceeds this range, extreme values ​​of 0.3 or 0.7 are taken respectively. For example, if the mask contour matching degree is 97%, the preset standard mask contour matching degree is 90%, and the first initial weight is 0.5, then the first optimized weight = (97% / 90%) × 0.5 = 0.54. Thus, the closer the mask contour matching degree is to the preset standard mask contour matching degree, the higher the calculated first optimized weight, reflecting the influence of individual adaptation differences.

[0090] Furthermore, the second optimization weight is obtained by subtracting the first optimization weight from 1. For example, if the first optimization weight is 0.54, then the second optimization weight = 1 - 0.54 = 0.46. The second optimization weight is inversely adapted to the first optimization weight. When the reliability of the average mask-face fit ratio in the visual dimension is insufficient, the second optimization weight in the pressure dimension is automatically increased.

[0091] Finally, based on the first and second optimization weights, the average first mask-to-face fit ratio and the average second mask-to-face fit ratio are weighted and calculated to obtain the mask-to-face fit degree. Wherein, mask-to-face fit degree = (first optimization weight × average first mask-to-face fit ratio) + (second optimization weight × average second mask-to-face fit ratio). For example, if the first optimization weight is 0.54, the second optimization weight is 0.46, the average first mask-to-face fit ratio is 80%, and the average second mask-to-face fit ratio is 75%, then the mask-to-face fit degree = (0.54 × 80%) + (0.46 × 75%) = 77.7%. The mask-to-face fit degree combines the average first mask-to-face fit ratio and the average second mask-to-face fit ratio; the higher the mask-to-face fit degree, the better the fit between the mask and the face.

[0092] In summary, compared to existing technologies, this application performs mask-face fit analysis on the user under monitoring based on the user's facial image sequence and mask-face pressure distribution sequence, outputting the mask-face fit degree. This achieves a quantitative assessment of mask-face fit degree, adapting to the complex operating scenarios of chemical workshops and providing reliable fit status data for subsequent failure risk prediction.

[0093] S50: Based on the set of workshop environmental parameter sequences, mask attribute information, and mask-face fit, predict the risk of mask failure of the gas mask, output the mask failure probability, and make a protection risk judgment and warning for the user to be monitored based on the mask failure probability.

[0094] The aforementioned steps obtain a set of workshop environmental parameter sequences, mask attribute information, and mask-face fit, which can be used to predict the probability of mask failure and provide a reliable quantitative basis for risk assessment and early warning.

[0095] To address the aforementioned issues, this application predicts the risk of gas mask failure based on the set of workshop environmental parameters, mask attribute information, and mask-face fit, outputs the mask failure probability, and assesses and warns the user under monitoring based on the mask failure probability.

[0096] Specifically, step S50 in the method includes:

[0097] A mask failure risk prediction plugin is constructed based on long short-term memory networks and convolutional neural networks. The mask failure risk prediction plugin includes J mask failure risk prediction models, where J is an integer greater than or equal to 5.

[0098] Based on the set of workshop environmental parameter sequences, environmental parameter fluctuation analysis is performed, and environmental parameter fluctuation coefficients are output.

[0099] Multiply the ratio of the environmental parameter fluctuation coefficient to the maximum workshop environmental parameter fluctuation coefficient within the historical time period by J and round it to obtain the optimization model selection quantity K, where K is greater than or equal to 1 and less than or equal to J;

[0100] K mask failure risk prediction models are randomly selected from the J mask failure risk prediction models. Based on the set of workshop environmental parameter sequences, mask attribute information and mask-face fit, the mask failure risk of the gas mask is predicted. The average of the K prediction results is calculated to obtain the mask failure probability.

[0101] In this embodiment, a mask failure risk prediction plugin containing J mask failure risk prediction models is first constructed based on a long short-term memory network and a convolutional neural network, where J is an integer greater than or equal to 5. In this way, the robustness of prediction is improved by combining multiple models to suit the complex environment of the chemical workshop.

[0102] Secondly, environmental parameter fluctuation analysis is performed based on the set of workshop environmental parameter sequences to quantify the severity of environmental parameter changes and output the environmental parameter fluctuation coefficient. The larger the environmental parameter fluctuation coefficient, the more severe the environmental parameter fluctuation, and the higher the uncertainty of failure risk. More mask failure risk prediction models are needed to participate in the prediction to reduce errors. Conversely, if the environmental parameters are stable, a small number of mask failure risk prediction models can meet the accuracy requirements.

[0103] Next, the ratio of the environmental parameter fluctuation coefficient to the maximum workshop environmental parameter fluctuation coefficient within the historical time period is multiplied by J and rounded to obtain the number of optimized models selected, K. The rounding includes at least rounding to the nearest integer, rounding up, and rounding down. Those skilled in the art can dynamically select the appropriate model based on the actual application scenario and prediction accuracy requirements. K is greater than or equal to 1 and less than or equal to J. For example, if the environmental parameter fluctuation coefficient is 0.112, the maximum workshop environmental parameter fluctuation coefficient within the historical time period is 0.6, and J is 5, rounding up yields K = ⌈(0.112 / 0.6) × 5⌉ = 1. This means only one mask failure risk prediction model is selected from the mask failure risk prediction plugin. This achieves a balance between prediction accuracy and prediction efficiency: fewer models are selected when environmental parameter fluctuations are small, reducing computational resource consumption; more models are selected when environmental parameter fluctuations are large, ensuring reliable prediction.

[0104] Finally, K mask failure risk prediction models are randomly selected from J mask failure risk prediction models. The set of workshop environmental parameters, mask attribute information, and mask-face fit are input into each of the K models, generating K prediction results. The average of these K results is then calculated to obtain the mask failure probability. For example, 3 mask failure risk prediction models are randomly selected from 5 models. These 3 models predict 3 results based on the set of workshop environmental parameters, mask attribute information, and mask-face fit: e.g., 0%, 82%, and 78%. The average of these results yields the mask failure probability = (80% + 82% + 78%) / 3 = 80%. This reduces the prediction error of a single model, improves the accuracy and robustness of the prediction, and provides a clear quantitative basis for subsequent risk assessment.

[0105] Furthermore, the risk assessment and early warning for monitored users can be based on the probability of mask failure. For example, multi-level warning thresholds can be set based on the potential consequences of mask failure, such as minor discomfort, moderate poisoning, or severe injury or death, referencing historical failure data. Differentiated responses can be matched to different levels. For instance, low risk is only recorded in the background; medium risk is alerted via smart bracelet and adjusted under the guidance of a safety officer; high risk triggers an audible and visual alarm and suspends work to replace equipment; and extremely high risk directly triggers emergency broadcasts and personnel evacuation. This ensures safety redundancy for high-risk operations while also maintaining production efficiency, shifting risk management from passive response to proactive prediction.

[0106] Specifically, the "mask failure risk prediction plugin based on long short-term memory network and convolutional neural network" includes:

[0107] An initial coupled model is constructed based on a long short-term memory network and a convolutional neural network;

[0108] Based on industrial big data, and guided by the protection of gas masks in chemical production workshops, a sample dataset was collected. The sample data includes a set of environmental parameter sequences of the sample workshop, sample mask attribute information, sample mask-face fit, and sample mask failure probability. The sample mask failure probability is the proportion of sample mask failure events under corresponding historical conditions.

[0109] The sample dataset is iteratively extracted with replacement to obtain J sample training sets. The initial coupled model is trained to convergence to obtain J mask failure risk prediction models, which are then combined to obtain a mask failure risk prediction plugin.

[0110] In this embodiment, since the mask failure risk prediction plugin needs to process both temporal data (such as workshop environmental parameter sequences) and structured data (such as mask attribute information and mask-face fit), a single network structure cannot handle both simultaneously. Therefore, an initial coupled model is constructed based on Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN). Exemplarily, the initial coupled model mainly consists of an LSTM branch, a CNN branch, and a fusion output layer. LSTM can capture the temporal dependencies of data, such as a continuous 30-minute increase in toxic gas concentration leading to accelerated saturation of the filter material and an accumulation of failure risk over time. CNN can transform discrete attribute features into high-dimensional correlated features through convolutional layers, such as a combination of poor mask filtering performance and low mask-face fit significantly increasing the failure risk. The fusion output layer concatenates the temporal features extracted by LSTM with the structured features extracted by CNN, and outputs the mask failure probability prediction value through a fully connected layer, achieving a comprehensive prediction of temporal trends and static attributes.

[0111] Secondly, based on industrial big data, and guided by the protection of gas masks in chemical production workshops, a sample dataset is collected. This sample data includes a set of environmental parameter sequences from the sample workshop, sample mask attribute information, sample mask-to-face fit, and sample mask failure probability. The sample mask failure probability represents the percentage of sample mask failure events under corresponding historical conditions. For example, based on industrial big data from chemical workshops, gas mask usage records, environmental monitoring data, and safety accident reports from the past 3-5 years are selected. Following the same methods and feature dimensions used to obtain the workshop environmental parameter sequences, mask attribute information, and mask-to-face fit, the set of environmental parameter sequences from the sample workshop, sample mask attribute information, and sample mask-to-face fit are extracted. The percentage of sample mask failure events under different historical conditions is then calculated to obtain the sample mask failure probability.

[0112] Finally, the sample dataset is iteratively extracted with replacement to obtain J sample training sets. The sample distribution of each training set is slightly different. Then, the initial coupled model is trained to convergence based on the J sample training sets to obtain J mask failure risk prediction models. These models are then combined to obtain the mask failure risk prediction plugin. For example, the mask failure risk prediction model can be trained through the following technical path: 1. Data preparation: Randomly select one sample training set from the J sample training sets and divide it into a training set, a validation set, and a test set in a ratio of 7:1.5:1.5. 2. Model Training: Using the sample workshop environment parameter sequence, sample mask attribute information, and sample mask-face fit as input features in the training set, and the corresponding sample mask failure probability as the supervision label, the Adam optimizer is used (initial learning rate 0.001, decaying by 10% every 5 rounds), with mean squared error (MSE) as the loss function. After each training round, the model's MSE and mean absolute error (MAE) are calculated using the validation set. When the change in MSE on the sub-validation set is <0.5% for 5 consecutive rounds, and the MAE on the test set is ≤2%, the model is considered converged, training is stopped, and one trained mask failure risk prediction model is obtained. Then, following the same method, J mask failure risk prediction models are trained to obtain J models, which are then combined to obtain a mask failure risk prediction plugin.

[0113] Specifically, the phrase "performing environmental parameter fluctuation analysis based on the set of workshop environmental parameter sequences and outputting environmental parameter fluctuation coefficients" includes:

[0114] Parameter fluctuation calculations are performed on multiple workshop environmental parameter sequences within the set of workshop environmental parameter sequences to obtain multiple parameter fluctuation values, wherein the parameter fluctuation value is the ratio of the parameter standard deviation to the parameter mean in the parameter sequence;

[0115] The environmental parameter fluctuation coefficient is calculated by weighting the fluctuation values ​​of the multiple parameters.

[0116] In this embodiment, parameter fluctuation calculations are first performed on multiple workshop environmental parameter sequences within the workshop environmental parameter sequence set to obtain multiple parameter fluctuation values. The parameter fluctuation value is the ratio of the parameter standard deviation to the parameter mean in the parameter sequence. The parameter standard deviation reflects the dispersion of the parameter, but it is affected by the magnitude of the parameter. Dividing by the parameter mean eliminates this magnitude effect. For example, the parameter standard deviation and parameter mean are calculated for each workshop environmental parameter sequence within the set. For instance, if a toxic gas concentration sequence within the set is [10ppm, 12ppm, 15ppm, 13ppm, 14ppm], the parameter mean = (10+12+15+13+14) / 5 = 12.8ppm, and the parameter standard deviation ≈ 1.92ppm, then the parameter fluctuation value = 1.92 / 12.8 = 0.15. Multiple parameter fluctuation values ​​are obtained using the same method, and these values ​​comprehensively reflect the fluctuation of the workshop environmental parameter sequence.

[0117] Secondly, the environmental parameter fluctuation coefficient is calculated by weighting the fluctuation values ​​of multiple parameters. For example, weights can be assigned to the parameter fluctuation values ​​based on the degree of influence of different workshop environmental parameters on the mask failure risk, where all weights sum to 1. For instance, based on prior data from the actual application scenario, weights of 0.4, 0.3, 0.1, 0.1, and 0.1 can be assigned to toxic gas composition and concentration, particulate matter composition and concentration, temperature, humidity, and air velocity, respectively. If the corresponding parameter fluctuation values ​​are 0.15, 0.12, 0.03, 0.05, and 0.08, then the weighted calculation yields the environmental parameter fluctuation coefficient = (0.15 × 0.4) + (0.12 × 0.3) + (0.03 × 0.1) + (0.05 × 0.1) + (0.08 × 0.1) = 0.112. The environmental parameter fluctuation coefficient comprehensively reflects the overall fluctuation of workshop environmental parameters; the smaller the environmental parameter fluctuation coefficient, the smaller the environmental fluctuation.

[0118] In summary, compared to existing technologies, this application predicts the mask failure risk of the gas mask based on the set of workshop environmental parameters, mask attribute information, and mask-face fit, outputs the mask failure probability, and assesses and warns the user under monitoring based on the mask failure probability. Thus, a multi-model mask failure risk prediction plugin addresses the complexity of chemical environments, and dynamically selects the number of models to balance prediction accuracy and efficiency, ultimately achieving reliable, efficient, and accurate failure risk prediction. This provides a scientific basis for assessing and warning of protective risks, reducing the probability of employees being exposed to hazardous environments.

[0119] In summary, the embodiments of this application have at least the following technical effects:

[0120] Compared to existing technologies, this application first monitors and acquires a set of workshop environmental parameter sequences within a preset time range in a chemical production workshop. This provides a reliable environmental data foundation for subsequent failure risk prediction.

[0121] Secondly, this application obtains the mask attribute information of the gas mask worn by the user to be monitored. In this way, a capability profile of the gas mask is obtained, providing crucial data on the mask's performance for subsequent prediction of mask failure probability.

[0122] Furthermore, this application acquires user facial image sequences of the user to be monitored using an industrial camera, and obtains the mask-face pressure distribution sequence of the gas mask using a pressure sensor. This provides a reliable data foundation for subsequent mask-face fit analysis.

[0123] Furthermore, this application performs mask-face fit analysis on the user to be monitored based on the user's facial image sequence and mask-face pressure distribution sequence, outputting the mask-face fit degree. This achieves a quantitative assessment of mask-face fit degree, adapting to the complex operating scenarios of chemical workshops and providing reliable fit status data for subsequent failure risk prediction.

[0124] Finally, this application predicts the mask failure risk of the gas mask based on the set of workshop environmental parameters, mask attribute information, and mask-face fit, outputs the mask failure probability, and assesses and warns the user under monitoring based on the mask failure probability. Thus, a multi-model mask failure risk prediction plugin addresses the complexity of chemical environments, and dynamically selects the number of models to balance prediction accuracy and efficiency, ultimately achieving reliable, efficient, and accurate failure risk prediction. This provides a scientific basis for assessing and warning of protective risks, reducing the probability of employees being exposed to hazardous environments.

[0125] Through the aforementioned technical solution, this application integrates workshop environmental parameter sequences, mask attribute information, user facial image sequences from a visual dimension, and mask-face pressure distribution sequences from a pressure dimension to achieve comprehensive perception of the protective status. Furthermore, based on a mask failure risk prediction plugin constructed using long short-term memory networks and convolutional neural networks, it predicts and outputs the mask failure probability, enabling real-time capture of risks such as environmental fluctuations, decreased mask-face fit, and performance degradation of gas masks. This improves the accuracy and reliability of protective monitoring, effectively reducing the exposure risk to workers and the probability of chemical safety accidents, providing dynamic and precise safety assurance for high-risk work scenarios.

[0126] Example 2, as Figure 2As shown, based on the same inventive concept as the AI-based intelligent wearable status monitoring method for protective equipment provided in Embodiment 1, this embodiment of the invention also provides an AI-based intelligent wearable status monitoring system for protective equipment, including:

[0127] The environmental parameter acquisition module 11 is used to monitor and acquire a set of workshop environmental parameter sequences within a preset time range in the chemical production workshop;

[0128] The mask attribute acquisition module 12 is used to acquire the mask attribute information of the gas mask worn by the user to be monitored;

[0129] Wearing a data acquisition module 13 is used to acquire a sequence of user facial images of the user to be monitored through an industrial camera, and to acquire a mask-face pressure distribution sequence of the gas mask through a pressure sensor;

[0130] The fit analysis module 14 is used to perform mask-face fit analysis on the user to be monitored based on the user's facial image sequence and mask-face pressure distribution sequence, and output the mask-face fit.

[0131] The risk warning module 15 is used to predict the risk of mask failure of the gas mask based on the set of workshop environmental parameter sequences, mask attribute information and mask-face fit, output the mask failure probability, and make a protection risk judgment and warning for the user to be monitored based on the mask failure probability.

[0132] The environmental parameter acquisition module 11 is specifically used for:

[0133] Configure environmental monitoring indicators, including the composition and concentration of toxic gases, the composition and concentration of particulate matter, temperature, humidity and air velocity;

[0134] Based on the environmental monitoring indicators, the environmental parameters of the chemical production workshop are monitored and obtained at multiple continuous monitoring time points within a preset time range, resulting in multiple workshop environmental parameter sequences, and a set of workshop environmental parameter sequences is constructed.

[0135] The mask attribute acquisition module 12 is specifically used for:

[0136] The mask attribute information of the gas mask includes at least the structural design features, filtration performance features, applicable environment features, and remaining service life.

[0137] The wearable data acquisition module 13 is specifically used for:

[0138] The system acquires a sequence of facial images of the user to be monitored using an industrial camera, and obtains a sequence of mask-face pressure distribution of the gas mask using a pressure sensor.

[0139] The fit analysis module 14 is specifically used for:

[0140] The user's facial images are acquired by an industrial camera at multiple consecutive monitoring time points within a preset time range, resulting in a sequence of user facial images.

[0141] The mask-face pressure distribution of the gas mask is obtained by monitoring multiple consecutive monitoring time points within a preset time range using pressure sensors, thus obtaining a mask-face pressure distribution sequence.

[0142] Based on the mask-face fit predictor, a first mask-face fit ratio sequence is predicted according to the user's facial image sequence;

[0143] The second mask-face fit ratio sequence was obtained by evaluating the mask-face pressure distribution sequence.

[0144] The mask-face fit is calculated based on the first mask-face fit ratio sequence and the second mask-face fit ratio sequence.

[0145] Specifically, the phrase "based on a mask-face fit predictor, predicting a first mask-face fit ratio sequence according to the user's facial image sequence" includes:

[0146] Based on historical wearing monitoring records of similar gas masks in chemical production workshops, a set of facial images of sample users was collected, and the mask-face fit ratio of different sample user facial images was labeled to obtain a set of sample fit ratio labels.

[0147] Using the set of sample user facial images as input and the set of sample fitting ratio labels as supervision, a convolutional neural network is trained until convergence to generate a mask-face fitting predictor.

[0148] Using the mask-face fit predictor, a first mask-face fit ratio sequence is predicted based on the user's facial image sequence.

[0149] Further, the step of "calculating the mask-face fit degree based on the first mask-face fit ratio sequence and the second mask-face fit ratio sequence" includes:

[0150] The mean values ​​of the first mask-face fitting ratio sequence and the second mask-face fitting ratio sequence are calculated respectively to obtain the mean value of the first mask-face fitting ratio and the mean value of the second mask-face fitting ratio.

[0151] The mask contour matching degree is calculated based on the facial physiological characteristics of the user to be monitored and the structural design features of the gas mask;

[0152] The ratio of the mask contour matching degree to the preset standard mask contour matching degree is set as the first weight adjustment coefficient, and multiplied by the first initial weight to obtain the first optimized weight, wherein the first initial weight is 0.5, and the first optimized weight is greater than or equal to 0.3 and less than or equal to 0.7.

[0153] The second optimization weight is obtained by subtracting the first optimization weight from 1.

[0154] Based on the first optimization weight and the second optimization weight, the average value of the first mask-face fit ratio and the average value of the second mask-face fit ratio are weighted and calculated to obtain the mask-face fit degree.

[0155] The risk warning module 15 is specifically used for:

[0156] A mask failure risk prediction plugin is constructed based on long short-term memory networks and convolutional neural networks. The mask failure risk prediction plugin includes J mask failure risk prediction models, where J is an integer greater than or equal to 5.

[0157] Based on the set of workshop environmental parameter sequences, environmental parameter fluctuation analysis is performed, and environmental parameter fluctuation coefficients are output.

[0158] Multiply the ratio of the environmental parameter fluctuation coefficient to the maximum workshop environmental parameter fluctuation coefficient within the historical time period by J and round it to obtain the optimization model selection quantity K, where K is greater than or equal to 1 and less than or equal to J;

[0159] K mask failure risk prediction models are randomly selected from the J mask failure risk prediction models. Based on the set of workshop environmental parameter sequences, mask attribute information and mask-face fit, the mask failure risk of the gas mask is predicted. The average of the K prediction results is calculated to obtain the mask failure probability.

[0160] Specifically, the "mask failure risk prediction plugin based on long short-term memory network and convolutional neural network" includes:

[0161] An initial coupled model is constructed based on a long short-term memory network and a convolutional neural network;

[0162] Based on industrial big data, and guided by the protection of gas masks in chemical production workshops, a sample dataset was collected. The sample data includes a set of environmental parameter sequences of the sample workshop, sample mask attribute information, sample mask-face fit, and sample mask failure probability. The sample mask failure probability is the proportion of sample mask failure events under corresponding historical conditions.

[0163] The sample dataset is iteratively extracted with replacement to obtain J sample training sets. The initial coupled model is trained to convergence to obtain J mask failure risk prediction models, which are then combined to obtain a mask failure risk prediction plugin.

[0164] Furthermore, the step of "conducting environmental parameter fluctuation analysis based on the set of workshop environmental parameter sequences and outputting environmental parameter fluctuation coefficients" includes:

[0165] Parameter fluctuation calculations are performed on multiple workshop environmental parameter sequences within the set of workshop environmental parameter sequences to obtain multiple parameter fluctuation values, wherein the parameter fluctuation value is the ratio of the parameter standard deviation to the parameter mean in the parameter sequence;

[0166] The environmental parameter fluctuation coefficient is calculated by weighting the fluctuation values ​​of the multiple parameters.

[0167] In summary, the embodiments of this application have at least the following technical effects:

[0168] Compared to existing technologies, this application firstly uses an environmental parameter acquisition module to monitor and acquire a set of workshop environmental parameter sequences within a preset time range, providing a reliable environmental data foundation for subsequent failure risk prediction. Secondly, through a mask attribute acquisition module, it acquires the mask attribute information of the gas mask worn by the user to be monitored, obtaining a self-capability profile of the gas mask, providing key data on the mask's performance for subsequent mask failure probability prediction. Thirdly, through a wear data acquisition module, it acquires user facial image sequences using an industrial camera and obtains mask-face pressure distribution sequences using a pressure sensor, providing a reliable data foundation for subsequent mask-face fit analysis. Furthermore, through a fit analysis module, it performs mask-face fit analysis on the user to be monitored based on the user facial image sequences and mask-face pressure distribution sequences, outputting the mask-face fit degree, realizing a quantitative assessment of mask-face fit degree, adapting to the complex operating scenarios of chemical workshops, and providing reliable fit status data for subsequent failure risk prediction. Finally, the risk warning module predicts the mask failure risk of the gas masks based on the set of workshop environmental parameter sequences, mask attribute information, and mask-face fit. It outputs the mask failure probability and, based on this probability, assesses and warns the protective risks to the monitored users. A multi-model mask failure risk prediction plugin addresses the complexity of chemical scenarios, and dynamically selects the number of models to balance prediction accuracy and efficiency. Ultimately, it achieves reliable, efficient, and accurate failure risk prediction, providing a scientific basis for protective risk assessment and early warning, and reducing the probability of employees being exposed to hazardous environments. This improves the accuracy and reliability of protective monitoring, effectively reduces the exposure risk to workers and the probability of chemical safety accidents, and provides dynamic and precise safety assurance for high-risk work scenarios.

[0169] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0170] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0171] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0172] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0173] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0174] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0175] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent wearable status monitoring of protective equipment based on AI, characterized in that the method... include: The system monitors and acquires a set of workshop environmental parameter sequences within a preset time range for chemical production workshops. Obtain the mask attribute information of the gas mask worn by the user to be monitored; The user's facial images are acquired by an industrial camera at multiple consecutive monitoring time points within a preset time range, resulting in a sequence of user facial images. The mask-face pressure distribution of the gas mask is obtained by monitoring multiple consecutive monitoring time points within a preset time range using pressure sensors, thus obtaining a mask-face pressure distribution sequence. Based on the mask-face fit predictor, a first mask-face fit ratio sequence is predicted according to the user's facial image sequence; The second mask-face fit ratio sequence was obtained by evaluating the mask-face pressure distribution sequence. The mask-face fit is calculated by weighting the first mask-face fit ratio sequence and the second mask-face fit ratio sequence. Based on the set of workshop environmental parameters, mask attribute information, and mask-face fit, the risk of mask failure of the gas mask is predicted by using a long short-term memory network and a convolutional neural network. The mask failure probability is output, and the protection risk is judged and warned for the user to be monitored based on the mask failure probability.

2. The method for intelligent wearable status monitoring of protective equipment based on AI according to claim 1, characterized in that, The monitoring acquires a set of workshop environmental parameter sequences within a preset time range for chemical production workshops, including: Configure environmental monitoring indicators, including the composition and concentration of toxic gases, the composition and concentration of particulate matter, temperature, humidity and air velocity; Based on the environmental monitoring indicators, the environmental parameters of the chemical production workshop are monitored and obtained at multiple continuous monitoring time points within a preset time range, resulting in multiple workshop environmental parameter sequences, and a set of workshop environmental parameter sequences is constructed.

3. The method for intelligent wearable status monitoring of protective equipment based on AI according to claim 1, characterized in that, The mask attribute information of the gas mask includes at least the structural design features, filtration performance features, applicable environment features, and remaining service life.

4. The method for intelligent wearable status monitoring of protective equipment based on AI according to claim 1, characterized in that, Based on the mask-face fit predictor, a first mask-face fit ratio sequence is predicted from the user's facial image sequence, including: Based on historical wearing monitoring records of similar gas masks in chemical production workshops, a set of facial images of sample users was collected, and the mask-face fit ratio of different sample user facial images was labeled to obtain a set of sample fit ratio labels. Using the set of sample user facial images as input and the set of sample fitting ratio labels as supervision, a convolutional neural network is trained until convergence to generate a mask-face fitting predictor. Using the mask-face fit predictor, a first mask-face fit ratio sequence is predicted based on the user's facial image sequence.

5. The method for intelligent wearable status monitoring of protective equipment based on AI according to claim 1, characterized in that, The mask-face fit is calculated based on the first mask-face fit ratio sequence and the second mask-face fit ratio sequence, including: The mean values ​​of the first mask-face fitting ratio sequence and the second mask-face fitting ratio sequence are calculated respectively to obtain the mean value of the first mask-face fitting ratio and the mean value of the second mask-face fitting ratio. The mask contour matching degree is calculated based on the facial physiological characteristics of the user to be monitored and the structural design features of the gas mask; The ratio of the mask contour matching degree to the preset standard mask contour matching degree is set as the first weight adjustment coefficient, and multiplied by the first initial weight to obtain the first optimized weight, wherein the first initial weight is 0.5, and the first optimized weight is greater than or equal to 0.3 and less than or equal to 0.

7. The second optimization weight is obtained by subtracting the first optimization weight from 1. Based on the first optimization weight and the second optimization weight, the average value of the first mask-face fit ratio and the average value of the second mask-face fit ratio are weighted and calculated to obtain the mask-face fit degree.

6. The method for intelligent wearable status monitoring of protective equipment based on AI according to claim 1, characterized in that, Based on the set of workshop environmental parameter sequences, mask attribute information, and mask-face fit, the risk of mask failure is predicted, and the mask failure probability is output, including: A mask failure risk prediction plugin is constructed based on long short-term memory networks and convolutional neural networks. The mask failure risk prediction plugin includes J mask failure risk prediction models, where J is an integer greater than or equal to 5. Based on the set of workshop environmental parameter sequences, environmental parameter fluctuation analysis is performed, and environmental parameter fluctuation coefficients are output. Multiply the ratio of the environmental parameter fluctuation coefficient to the maximum workshop environmental parameter fluctuation coefficient within the historical time period by J and round it to obtain the optimization model selection quantity K, where K is greater than or equal to 1 and less than or equal to J; K mask failure risk prediction models are randomly selected from the J mask failure risk prediction models. Based on the set of workshop environmental parameter sequences, mask attribute information and mask-face fit, the mask failure risk of the gas mask is predicted. The average of the K prediction results is calculated to obtain the mask failure probability.

7. The method for intelligent wearable status monitoring of protective equipment based on AI according to claim 6, characterized in that, A mask failure risk prediction plugin is built based on Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks, including: An initial coupled model is constructed based on a long short-term memory network and a convolutional neural network; Based on industrial big data, and guided by the protection of gas masks in chemical production workshops, a sample dataset was collected. The sample data includes a set of environmental parameter sequences of the sample workshop, sample mask attribute information, sample mask-face fit, and sample mask failure probability. The sample mask failure probability is the proportion of sample mask failure events under corresponding historical conditions. The sample dataset is iteratively extracted with replacement to obtain J sample training sets. The initial coupled model is trained to convergence to obtain J mask failure risk prediction models, which are then combined to obtain a mask failure risk prediction plugin.

8. The method for intelligent wearable status monitoring of protective equipment based on AI according to claim 6, characterized in that, Based on the aforementioned set of workshop environmental parameter sequences, environmental parameter fluctuation analysis is performed, and environmental parameter fluctuation coefficients are output, including: Parameter fluctuation calculations are performed on multiple workshop environmental parameter sequences within the set of workshop environmental parameter sequences to obtain multiple parameter fluctuation values, wherein the parameter fluctuation value is the ratio of the parameter standard deviation to the parameter mean in the parameter sequence; The environmental parameter fluctuation coefficient is calculated by weighting the fluctuation values ​​of the multiple parameters.

9. An AI-based intelligent wearable status monitoring system for protective equipment, characterized in that, For performing the method according to any one of claims 1-8, comprising: The environmental parameter acquisition module is used to monitor and acquire a set of workshop environmental parameter sequences within a preset time range in the chemical production workshop; The mask attribute acquisition module is used to acquire the mask attribute information of the gas mask worn by the user to be monitored; The user face image acquisition module is used to acquire user face images of the user to be monitored at multiple consecutive monitoring time points within a preset time range through an industrial camera, and obtain a user face image sequence. The mask-face pressure distribution monitoring module is used to monitor and acquire the mask-face pressure distribution of the gas mask at multiple consecutive monitoring time points within a preset time range through a pressure sensor, and obtain a mask-face pressure distribution sequence. The first mask-face fit ratio prediction module is used to predict the first mask-face fit ratio sequence based on the user's facial image sequence using the mask-face fit predictor. The second mask-face fit ratio evaluation module is used to evaluate and obtain the second mask-face fit ratio sequence based on the mask-face pressure distribution sequence. The mask-face fit calculation module is used to calculate the mask-face fit by weighting the first mask-face fit ratio sequence and the second mask-face fit ratio sequence. The risk warning module is used to predict the risk of mask failure of the gas mask by using a long short-term memory network and a convolutional neural network based on the set of workshop environmental parameter sequences, mask attribute information and mask-face fit, output the mask failure probability, and make a protection risk judgment and warning for the user to be monitored based on the mask failure probability.

Images